Depot 42 Signal — how we built an AI radar for projects in 7 days

We wanted a way to see the real state of every project at once.

The problem wasn't a lack of data. On the contrary, we had too much of it: messages, comments, status updates, questions, calls. The information existed, but it was scattered across tools and buried in day-to-day communication.

To understand what was happening on a project, leadership had to piece the picture together manually: read messages, attend meetings, recall previous decisions, and connect everything to deadlines, budgets, and client sentiment. That took at least an hour every day.

For a single project, that still works. Across several projects, blind spots start to appear. That's usually where the most expensive problems begin.

We wanted to bring data from different sources into one place and automatically surface both risk signals and positive signals that could affect a project's progress. The goal was to reduce the need for manual monitoring and save time for managers and company leaders.

We could have hired an operations director to keep track of project status across the company. Instead, we decided to build a system that shows the real state of all projects at once. We wanted to create a working MVP without a development team, using vibe coding and modern AI tools.

That's how Depot 42 Signal appeared — an AI-powered service for analyzing project context. It helps leaders and project managers spot signals that influence how projects move forward.

We built Brele Signal using vibe coding.

There was no traditional development process: no lengthy specifications, no engineering team, and no dedicated development budget. What we did have was a standard product iteration cycle: idea → prototype → testing → insights → next iteration.

Depot 42 CEO, Andrey, defined the product logic, tested user flows and interfaces, and continuously refined the system's behavior.

AI wasn't just generating code. It acted as a technical partner. Andrey described what he wanted to achieve, reviewed the output, provided feedback, and refined the requirements through multiple iterations.

Within a week — without stepping away from day-to-day operations — we had a working internal tool that included:

• Projects.
• Signals.
• Meetings.
• Chats.
• Dashboards.
• A Kanban board for tasks.
• An admin panel.
• Project health tracking.
• Two languages.
• Light and dark themes.
• AI prompt configuration.

In essence, we built a system that can be adapted to the workflows of almost any agency.

Not long ago, an idea like this would have automatically become a mid-sized software project requiring a dedicated development team. Today, a founder or manager who understands the problem and the product can build the first working version themselves.

From the outside, it may look like everything is generated "at the click of a button" with AI. In reality, AI is an accelerator, not a magic wand. Knowing how to work with AI matters. But even more important is the ability to validate hypotheses, design systems, define business logic, test user scenarios, identify flaws, and make architectural decisions.

Before becoming Depot 42 Signal, the product went through dozens of iterations and evolved step by step.

The first step was bringing together information that normally lives in separate tools:

• Meeting recordings and transcripts, automatically imported through email integrations.
• Project-related chats, both internal and client-facing.
• Manager comments and notes within the system.
• Client NPS scores.

Depot 42 Signal connects to these sources, analyzes communication patterns, and highlights meaningful signals. For example, it can identify delivery risks, signs of client dissatisfaction, team fatigue, or, on the positive side, growing confidence, positive dynamics, and overall project stability.

We designed the system so the AI works within clear boundaries. To achieve this, we created a controlled signal framework that can be configured through the admin panel.

Signal types include:

• Delivery Risk.
• Client Dissatisfaction.
• Team Demotivation.
• Budget Risk.
• Client Satisfaction.
• Communication Breakdown.

The AI analyzes project context and maps events to specific signal types.

Each signal has its own weight and impact on project health metrics, allowing the system to evaluate not only what is happening but also how important it is.

We evaluate project status using three metrics that matter most to us:

• Project Health.
• Client Satisfaction.
• Team Engagement.

Based on these metrics, we built dashboards that provide an instant overview of what's happening across projects and highlight where attention is needed.

This replaced the need to constantly scan chats, meeting notes, and discussions by hand.

The main dashboard gives us a high-level view of:

• Active projects.
• Overall project health.
• Risks.
• Key signals.

To avoid losing important agreements made in chats and meetings, we built a client task board inside Signal.

Every day, the system analyses conversations and meeting transcripts, then automatically creates tasks based on the context. For example, a client asks a question and the team promises to follow up. Or additional information is required from the client to move work forward.

Instead of relying on someone to manually capture these action items, Signal turns them into tasks automatically and places them on the board.

We also brought everything that influences system behaviour into a single admin panel:

• Analysis logic.
• AI prompts.
• Metric calculation rules.
• Signal processing workflows.
• AI model configuration.

As a result, Signal became much easier to manage. We can adjust signal weights, refine prompts, switch AI models, and fine-tune system behaviour without changing the product itself.

We integrated NPS to collect client feedback at the end of every sprint. This gives us a consistent measure of how clients evaluate project progress and helps track changes over time.

It also provides a useful counterbalance to individual signals. Rather than reacting to every positive or negative event that happens during the week, we can see the broader trend in client satisfaction.

We also created a support bot called Sig Forty-Two. The name comes from Depot 42.

Bot politely reminds clients to complete NPS surveys, sits in project chats, generates summaries, and helps capture signals that emerge throughout the project lifecycle.